Study Design, Part 2 — Self-Controlled and Specialised Designs

Beyond the two foundational designs Lesson 15.4 covered, pharmacoepidemiology has developed several more specialised designs, each solving a specific methodological problem the basic cohort and case-control approaches handle poorly. The Self-Controlled Case Series (SCCS) is arguably the most elegant of these: it uses each patient as their own control, comparing that individual patient’s risk of the adverse event during drug-exposed periods against the same patient’s risk during their own unexposed periods. Because every comparison happens within one patient, SCCS eliminates all time-stable between-patient confounding entirely — genetics, chronic comorbidities, socioeconomic factors, anything that doesn’t change over the study period simply cancels out by design, without needing to be measured or adjusted for at all.

That elegance comes with one absolutely critical assumption that, if violated, breaks the design completely: the adverse event must not itself affect the patient’s future drug exposure. If an event is severe enough that the drug gets discontinued immediately afterward — which is exactly what happens with a serious ADR — then exposure and outcome are no longer independent of each other, and SCCS’s core statistical logic no longer holds. This is precisely why the worked example in this lesson’s callout walks through choosing an active comparator cohort design instead for a Drug Y and pancreatitis signal — pancreatitis is exactly the kind of event that gets a drug stopped immediately, which disqualifies SCCS despite its methodological attractiveness for acute-onset events.

Nested case-control designs combine cohort and case-control efficiency directly: within a large, already-defined cohort study, cases get fully validated and matched controls are sampled from within that same cohort, rather than validating every single outcome for the entire cohort population — a substantial efficiency gain when outcome validation (confirming a case genuinely meets the diagnostic criteria) is expensive. The critical methodological care required is in control selection — controls have to be sampled to preserve the actual ratio of exposed to unexposed patients at the specific time each case occurred, a technique called risk-set sampling, or the resulting comparison becomes distorted.

Sequence Symmetry Analysis (SSA) is a genuinely clever, specialised tool for one specific scenario: detecting when a suspected adverse drug reaction leads to a new prescription for another drug used to treat it. It compares the sequence of drug initiation — do more patients start the second drug after beginning the first than before it — with symmetrical initiation patterns suggesting no association and asymmetrical patterns suggesting a genuine causal signal, all without needing a separate unexposed control group at all. And Interrupted Time Series (ITS) is the one design in this lesson built for a fundamentally different purpose than detecting a new signal: it evaluates whether an already-taken regulatory action — a label change, a safety communication, a Dear Healthcare Professional letter — actually changed prescribing or outcome patterns at the population level, by comparing the trend before the intervention against the trend after it. This closes a loop this course has referenced since Module 3: it’s the methodology that actually measures whether the six-stage signal-to-label feedback loop produced a real-world effect, rather than just assuming a label change worked.